feat(alibaba): add scoped Qwen reasoning options

[rekram1-node]

Summary

• add normalized toggle and budget_tokens reasoning options to a deliberately scoped Alibaba/DashScope Qwen matrix
• cover both represented pay-as-you-go OpenAI-compatible Chat Completions regional surfaces: alibaba (international) and alibaba-cn (Chinese mainland)
• keep unsupported or insufficiently verified endpoint contracts out of this PR

Verified records

Provider surface	Model	Normalized controls	Verified wire controls
alibaba	qwen-plus	toggle, budget_tokens	enable_thinking, thinking_budget
alibaba	qwen3.5-plus	toggle, budget_tokens	enable_thinking, thinking_budget
alibaba	qwen3-vl-plus	toggle, budget_tokens	enable_thinking, thinking_budget
alibaba-cn	qwen-plus	toggle, budget_tokens	enable_thinking, thinking_budget
alibaba-cn	qwen3.5-plus	toggle, budget_tokens	enable_thinking, thinking_budget
alibaba-cn	qwen3-vl-plus	toggle, budget_tokens	enable_thinking, thinking_budget

Official sources

• Chat Completions deep-thinking contract and regional URL examples: https://www.alibabacloud.com/help/en/model-studio/deep-thinking
• Chinese-mainland Chat Completions deep-thinking contract and regional URL examples: https://help.aliyun.com/zh/model-studio/deep-thinking
• Multimodal Qwen thinking controls, including qwen3.5-plus and qwen3-vl-plus: https://www.alibabacloud.com/help/en/model-studio/vision
• Chinese-mainland multimodal Qwen thinking controls: https://help.aliyun.com/zh/model-studio/vision
• Responses API contract, audited separately because it introduces Responses-only reasoning.effort: https://help.aliyun.com/zh/model-studio/qwen-api-via-openai-responses
• Token Plan contract audited separately: https://www.alibabacloud.com/help/en/model-studio/token-plan-overview
• Coding Plan contracts audited separately: https://www.alibabacloud.com/help/en/model-studio/coding-plan and https://help.aliyun.com/zh/model-studio/coding-plan

Wire-to-normalized mapping

• Chat Completions enable_thinking: true|false maps to normalized toggle.
• Chat Completions thinking_budget maps to normalized budget_tokens; it is a dedicated ceiling for reasoning tokens.
• Responses API reasoning.effort was not added because these provider records represent the OpenAI-compatible Chat Completions endpoint contract, not a Responses-only surface.

Verification boundaries and exclusions

• This is intentionally not a model-identity propagation patch. Each included record has an exact first-party Chat Completions example or an exact multimodal family contract on both regional endpoint surfaces.
• Fixed-thinking models such as qwq-plus, qwen3-next-80b-a3b-thinking, and dedicated thinking-suffix variants are excluded because callers cannot switch reasoning off.
• Other Alibaba and Alibaba China Qwen records, including newer Qwen3.6/Qwen3.7 rows, open-source families, QVQ, coder, and omni records, are excluded from this safe first PR where the audited official docs did not directly verify the exact represented endpoint/model contract to the same standard.
• alibaba-token-plan is excluded: the official Token Plan overview establishes a separate subscription surface but does not directly document caller-selectable enable_thinking/thinking_budget controls for the represented international endpoint records.
• alibaba-coding-plan and alibaba-coding-plan-cn are excluded: these are separate plan-specific endpoint contracts. The plan docs and Qwen Code setup examples are not treated as sufficient proof of the normalized caller-selectable matrix for every represented record.
• No normalized effort options are added: audited effort values belong to the Responses API surface.

Validation

• bun install --frozen-lockfile in the isolated worktree (dependencies only; no tracked file changes)
• bun validate
• git diff --check